Journal of Hydrology 540 (2016) 64–81

Contents lists available at ScienceDirect

Journal of Hydrology

journal homepage: www.elsevier .com/locate / jhydrol

Review Paper

Regional scale hydrologic modeling of a karst-dominant geomorphology:The case study of the Island of Crete

http://dx.doi.org/10.1016/j.jhydrol.2016.05.0610022-1694/� 2016 European Commission, Joint Research Centre. Published by Elsevier B.V.This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).

⇑ Corresponding author at: European Commission, Joint Research Centre, Institute for Environment and Sustainability, Ispra, VA, Italy.E-mail address: ing.anna_malago@libero.it (A. Malagò).

Anna Malagò a,c,⇑, Dionissios Efstathiou b, Fayçal Bouraoui a, Nikolaos P. Nikolaidis b, Marco Franchini c,Giovanni Bidoglio a, Marinos Kritsotakis d

a European Commission, Joint Research Centre, Institute for Environment and Sustainability, Ispra, VA, Italyb Technical University of Crete, 73100 Chania, Crete, GreececUniversity of Ferrara, Engineering Department, Ferrara, ItalydDecentralized Administration of Crete, Directorate of Water, 71202 Heraklion, Greece

a r t i c l e i n f o s u m m a r y

Article history:Received 14 October 2015Received in revised form 22 May 2016Accepted 24 May 2016Available online 30 May 2016This manuscript was handled by CorradoCorradini, Editor-in-Chief, with theassistance of Okke Batelaan, AssociateEditor

Keywords:SWATKarstMediterraneanWater balanceMulti-site calibrationCrete

Crete Island (Greece) is a karst dominated region that faces limited water supply and increased seasonaldemand, especially during summer for agricultural and touristic uses. In addition, due to the mountain-ous terrain, interbasin water transfer is very limited. The resulting water imbalance requires a correctquantification of available water resources in view of developing appropriate management plans to facethe problem of water shortage.The aim of this work is the development of a methodology using the SWAT model and a karst-flow

model (KSWAT, Karst SWAT model) for the quantification of a spatially and temporally explicit hydrologicwater balance of karst-dominated geomorphology in order to assess the sustainability of the actual wateruse. The application was conducted in the Island of Crete using both hard (long time series of streamflowand spring monitoring stations) and soft data (i.e. literature information of individual processes). TheKSWAT model estimated the water balance under normal hydrological condition as follows:6400 Mm3/y of precipitation, of which 40% (2500 Mm3/y) was lost through evapotranspiration, 5% wassurface runoff and 55% percolated into the soil contributing to lateral flow (2%), and recharging the shal-low (9%) and deep aquifer (44%). The water yield was estimated as 22% of precipitation, of which abouthalf was the contribution from spring discharges (9% of precipitation). The application of the KSWATmodel increased our knowledge about water resources availability and distribution in Crete under differ-ent hydrologic conditions. The model was able to capture the hydrology of the karst areas allowing a bet-ter management and planning of water resources under scarcity.� 2016 European Commission, Joint Research Centre. Published by Elsevier B.V. This is an open access

article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).

Contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 652. Materials and methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

2.1. Study area . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 652.2. The karst geology and hydrology of Island of Crete . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 662.3. Streamflow and springs measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 672.4. Approach for the prediction of karst water resources. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

2.4.1. The SWAT model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 672.4.2. Adapting SWAT model to simulate karst processes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 682.4.3. Model setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 702.4.4. Calibration of streamflow. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 712.4.5. The regionalization and classification of ungauged subbasins. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

A. Malagò et al. / Journal of Hydrology 540 (2016) 64–81 65

2.4.6. The calibration of springs and delineation of karst recharge area of each spring . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 732.4.7. Final calibration and complete validation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73

3. Results and discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

3.1. Hydrological simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 743.2. The estimated recharge areas of karst springs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 753.3. Spatial and temporal variation of hydrological components in Crete . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 773.4. The estimated water balance of Crete . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 773.5. Strengths and challenges of the karst modeling approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

4. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

1. Introduction

‘‘Karst” identifies a specific geological landscape and morphol-ogy formed by the dissolving action of water on soluble carbonaterocks such as primarily limestone, but also marble, dolomite, andgypsum. These rocks are mechanically strong but chemically sol-uble with high degree of secondary porosity. As a consequence,the hydrological cycle provides the primary source of energy forkarst formation because water is the solvent that dissolves carbon-ate rocks and then carries the ions away in solution (Williams,2004).

The process of dissolution (‘karstification’) leads to the develop-ment of caves, sinkholes, springs and sinking streams that are typ-ical features of a karst system. With progressing karstification,groundwater flow in the karst aquifer develops from a flow in aninterconnected fissure network to a flow concentrated in severallarge pipes, interconnected cavities and cave systems (EC, 2003,2004). The downstream end of a karst system usually is a springwhere the underground conduit reaches the surface as an outputpoint from an extensive network of groundwater conduits (Smartand Worthington, 2004).

In Europe, soluble carbonate rocks are widespread in Western,Southern and Eastern part covering 35% of whole Europe (Dalyet al., 2002), so that the karst processes are significant componentsof the physical geography of Mediterranean basins. In particular,limestones reach great thickness in Spain, southern France, Italy,the Balkan Peninsula, Turkey and in many islands in the Mediter-ranean (Crete, Majorca and Sicily). As a consequence, karst aquifersand springs are an important source of water supply for Mediter-ranean countries and special strategies are required to managethe quantity and quality of their waters.

Bakalowicz (2015) pointed out the importance to study thekarst aquifer functioning and the local geological evolution in orderto manage in realistic and sustainable way the water resources.The monitoring and management of these resources are recognizedin Europe as an essential issue and the European Union promptedthe creation of COST Actions 620 and 621 to develop a comprehen-sive methodology for risk assessment and for the sustainable man-agement of karst systems (EC, 2003, 2004). Unfortunately, mostcountries are lacking behind in monitoring the discharge of springsor wells and the exploitation of karst aquifers in generally is inap-propriate (Bakalowicz, 2015).

In this context, large scale hydrologic models are essential toolsfor watershed management at regional scale. Regional scale mod-els with an appropriate discretization of watershed can adequatelyaccount for the spatial heterogeneity improving water predictions(Wooldridge and Kalma, 2001). A variety of karst models have beendeveloped and applied to karst watersheds (Nikolaidis et al., 2013).Recently, Hartmann et al. (2015) presented for the first time thesimulations of groundwater recharge in Europe with a grid-hydrological model (VarKarst-R) pointing out the importance of acharacterization of subsurface heterogeneity.

Baffaut and Benson (2009) modified the SWAT (Soil and WaterAssessment Tool) model to simulate faster aquifer recharge in aMissouri karst watershed (SWAT-B&B) modifying deep groundwa-ter recharge equations, increasing the hydraulic conductivity ofsinkholes simulated as ponds and losses from streams. After that,Yachtao (2009) further modified the SWAT model in order toimprove the simulation of water quality and quantity in the Ope-quon Creek watershed (USA). The author introduced two newparameters for simulating the hydrology and nitrate transport ina sinkhole and losses from sinking streams. More recently, Wangand Brubaker (2014) proposed a non-linear modification ofgroundwater algorithm in SWAT (ISWAT) improving the recessionand low-flow simulation. Nikolaidis et al. (2013) developed a reser-voir model approach linked with SWAT to simulate karst’s behav-ior and the recharge of springs of Koiliaris basin in Crete addingfive new parameters.

All these studies indicate that a specific parameterization ofaquifer discharge, return flow, stream losses and sinkholes isrequired in karst watersheds. In addition, Wang et al. (2014)showed that a specific database of karst SWAT soils should be usedto assess the influence of soil hydrological process in a karst region.

In this paper the SWAT model was integrated with a karst-flowmodel (Nikolaidis et al., 2013; Tzoraki and Nikolaidis, 2007). Karst-subbasins were defined in order to reproduce more accurately thewater balance at regional scale.

This integrated model, KSWAT model, was applied to the Islandof Crete, one of the most intensively managed Mediterraneanislands, where the major water use is irrigation (84.5% of the totalconsumption) and the main water source is karst aquifers. Thegrowing water demand of the region makes the rational manage-ment of water resources extremely important in view of a sustain-able development. Consequently, the specific objective of thisstudy is the development of a large scale methodology for thequantification of a spatially explicit water balance of karst-dominated geomorphologies using the SWAT model in order toassess the sustainability of water use in the Island of Crete andpotentially to other areas.

2. Materials and methods

2.1. Study area

The Island of Crete occupies the southern part of Greece and isthe largest island of Greece and the fifth in the Mediterranean. TheIsland covers an area of 8336 km2 and is divided into prefectures,including from east to west: Lasithi (1810 km2), Heraklion(2626 km2), Rethymno (1487 km2) and Chania (2342 km2)(Fig. 1a). The maximum length of the Island is 269 km and themaximum width 60 km. Four main mountains run west to east:the White Mountains in the west (2453 m), Idis mountain(2456 m) in the center, Asterousian (1280 m) in south Heraklionand Dikti (2148 m) in the east (Baltas and Tzoraki, 2013).

Fig. 1. Map of prefectures (a); the main carbonate rocks of Crete (karst areas from geological map) with springs and streamflow gauging stations (b) A, White Mountainous; B,Idi; C, Dikti; D, Sitia; map of the main river basins (c).

Table 1Description of available streamflow and spring discharges measurements in the Crete Island provided by the bureau of Water Resources at the Decentralized Administration ofthe Region of Crete. The symbol # represent the number.

Type of measure # gaugingstations

Type oflocation

Type ofmeasures

Frequency Period extension # Calibratedgauging stations

# Validatedgaugingstations

Streamflow 22 Permanentgaugingstation

Waterdepth

Continuous measurements on agraphic record (Rantz et al., 1982)

1980–2009 (30 years;#5120 data entries)

15 7

Spring Discharge 47 Permanentgaugingstation

Watervelocity

Once a month 1983–2005 (23 years;#8015 data entries)

47 –

66 A. Malagò et al. / Journal of Hydrology 540 (2016) 64–81

The Island of Crete is characterized by a dry semi-humidMediterranean climate with dry and warm summers and humidand relatively cold winters where mean annual rainfall decreasesfrom west to east and from north to south, but increases with alti-tude (MEDIWAT, 2013). Annual precipitations are highly variableranging between 300 mm in coastal areas and 2000 mm in head-waters in White Mountains. The mean annual temperature rangesfrom 18.5� in the west to 20� in the south of island and decreaseswith altitude.

The mountainous areas, in particular in the western part, havemountainous climate. As a consequence, Crete contains sub-regions with very different hydrological characteristics.

Crete has about 2550 km2 of agriculture land, about 30% ofwhole Crete, with more than 1100 km2 in Heraklion, and

3800 km2 of pasture (45% of total areas of Crete). The main cropsare olives, grapes, and the main vegetables crops are tomatoes,cucumbers, onions, potatoes, watermelons and melons. Thedemand for irrigation water is high (about 360 Mm3/y), while only47% (1200 km2) of agricultural land is irrigated. In Heraklion theirrigated area is around 600 km2, followed by Lasithi and Chaniawith around 300 km2, while a very small area in Rethymno is irri-gated (Agriculture statistics of Greece, 2005).

2.2. The karst geology and hydrology of Island of Crete

The geology of Crete is composed of carbonate rocks (limestone,marble and dolomite) which allow water to penetrate, creatingmajor karst formation (Baltas and Tzoraki, 2013). More than 30%

Fig. 2. Karst SWAT modeling flowchart.

A. Malagò et al. / Journal of Hydrology 540 (2016) 64–81 67

of carbonate rocks cover the total area of the Island and the majorcarbonate rocks of Crete are located in the White mountains (LefkaOri), Idi, Dikti and Sitia (Fig. 1b). The total karst area covers about2730 km2 and the water contribution to the karst aquifer is esti-mated around 2000 Mm3/y which discharges out in many springs(Chartzoulakis et al., 2001).

There are 47 gauged springs in Crete with relative large flow,which are subdivided in three main classes: freshwater springs,brackish water springs and undersea springs. Most of springs arekarst springs and refer to the same karst hydrogeological system(Lefka Ori, Idi, Dikti and Sitia) discharging around 500 Mm3/yfreshwater into rivers. The most important springs are Stylos, Pla-tanos and Kourtaliotis that discharge 85 Mm3/y in the Kalami/Koil-iaris basin, 67 Mm3/y in Plantanias Basin and 38 Mm3/y inKourtaliotis basin, respectively (Fig. 1c).

Big brackish springs, located on coastal areas, include Almiros-Heraklion that discharges around 235 Mm3/y, Almiros-Agios-Nikolaos (83 Mm3/y) and brackish springs of lower dischargeinclude Almiros-Mallia (3.2 Mm3/y), Georgioupolis, Grammatikakiand Malavra. Brackish springs discharge directly about 285 Mm3/yin the sea. Submarine discharges can be found in the southern partof Lefka Ori, Souda bay, Bali bay, Mallia bay, Elounta, Skinia, andthe eastern part of the limestones of Zakros (MEDIWAT, 2013).

2.3. Streamflow and springs measurements

Stream and spring discharge measurements were provided bythe bureau of Water Resources at the Decentralized Administrationof the Region of Crete that deals with data elaboration and theassessment of their accuracy. The stream measurements at 25 per-manent locations were conducted using a continuous graphicrecord of the water depth. For a comprehensive description of con-tinuous measurements of water depth, the reader can refer toRantz et al. (1982).

These measurements were then converted to streamflow ratewith an established rating curve (streamflow, m3/s vs water depth,m) calculated every year for each monitoring station following theprocedure described in Buchanan and Somers (1976). Thestreamflow-water depth relationship was adopted to calculatethe streamflow at daily time step, and then aggregated at monthlytime step. The available dataset for this study included 22 gaugingstations because the streamflow stations STR30 and STR29 wereexcluded due to their uncertain localization and drain areas.STR5 was also excluded because observed data were lacking duringthe study period 1980–2009.

The spring’s discharges measurements were directly calculatedby measuring velocity once a month on surveyed cross-sectionalchannel geometry at 47 permanent locations. The only exceptionwas the Almiros spring for which only the calculated monthly dis-charge was available. Table 1 summarizes the streamflow andsprings data utilized in this study.

2.4. Approach for the prediction of karst water resources

The proposed karst modeling approach links the process-basedmodel SWAT and a karst-flow model (Nikolaidis et al., 2013). Fig. 2represents the sequential steps of our approach that involved thesetup of SWAT model, the identification of karst subbasins inwhich a modified version of SWAT model was applied, the calibra-tion of streamflow at selected gauging stations (step-wise calibra-tion), the regionalization of calibrated parameters for ungaugedsubbasins, the spring’s discharges calibration outside the SWATmodel using the karst-flow model and finally the introduction ofcalibrated discharge of springs as a point sources in the SWATmodel. The overall procedure included the use of hard and softdata. Hard data are defined as long-term, measured time series,

typically at a point within a watershed, while soft data are definedas information on individual processes within a budget that maynot be directly measured within the study area, generally from lit-erature information (Arnold et al., 2015). In this study long timeseries of 22 streamflow and 47 of spring monitoring stations wereused as hard data, while for instance literature information of theextension of karst recharge areas, as soft data. This approachenables the dialog between local experts, experimentalists andmodelers, increases the reliability of the model results in ungaugedareas and helps to better constrain model parameters (Seibert andMcDonnell, 2002).

2.4.1. The SWAT modelThe SWAT model (Arnold et al., 1998) model is a continuous-

time, semi-distributed, process based river basin model, originallydeveloped to predict the long-term impact of climate and land usemanagement practices on water, sediment, and agricultural

Fig. 3. Comparison between the classic SWAT model configuration and the adapted SWAT model linked with the karst-flow model. In (a): DA_RCHRG, the deep aquiferrecharge; ET, evapotranspiration; GW_RCHRG, the shallow aquifer recharge; GWQ1, baseflow from shallow aquifer; GWQ2, baseflow from deep aquifer; INF, infiltration in thesoil; LATQ, Lateral flow; SURQ, surface runoff. In (b): a1, fraction of DA_RCHRG to the upper reservoir; a2, fraction of flow from upper to lower reservoir; DA_RCHRG, amount ofdirect recharge of deep aquifer from several subbasins (inlet of the karst-flow model); ET, evapotranspiration; INF, infiltration in the soil; Q1, outlet of upper reservoir; Q2,outlet of lower reservoir; Qk, calibrated spring’s discharge; SURQ, surface runoff; SURQN, surface runoff excluding TLOSS and TWLWET; TLOSS, tributary stream losses;TWLWET, losses from the bed of wetlands.

68 A. Malagò et al. / Journal of Hydrology 540 (2016) 64–81

chemical yields in large complex basins. SWAT uses the water bal-ance approach to simulate watershed hydrologic partitioning. Inparticular, the water balance is calculated considering the snow,soil, shallow aquifer and deep aquifer components.

The watershed is first subdivided into subbasins and each sub-basin into hydrologic response units (HRUs) that are a function ofsoil type, land use and land slope. The main components of themodel are: weather, hydrologic mass balance, soil temperatureand soil properties, plant growth, nutrients cycling, and sedimentsyield. SWAT runs on a daily time step. For specific aspects of SWAT,Neitsch et al. (2010) provide a complete description in the theoret-ical documentation of the model.

2.4.2. Adapting SWAT model to simulate karst processesSWAT2012 (v.622) uses two linear reservoirs to partition

groundwater into two aquifer systems (Fig. 3a): a shallow aquiferwhich contributes baseflow to streams (GWQ1) and a deep aquiferwhich can also contribute baseflow to streams (GWQ2). Theremaining portion in the deep aquifer (DA_RCHRG – GWQ2) canbe considered lost from the system. The recharge for a specificday is calculated as a linear function of the daily seepage, therecharge of the previous day and the groundwater delay. Dailyseepage includes seepage through the soil profile, through pondsor wetlands and loosing streams (tributaries and main channels).All seepage losses are added together and assumed to travel verti-cally to the aquifer with the same velocity. The baseflow from theshallow aquifer GWQ1 is calculated through the aquifer recharge(GW_RCHRG, mm H2O), the deep aquifer recharge (DA_RCHRG,mm H2O) and the baseflow recession constant (ALPHA_BF,1/day). Similarly, the baseflow from the deep aquifer (GWQ2)depends on the deep aquifer recharge and on the deep-baseflowrecession constant (ALPHA_BF_D, 1/day).

In order to simulate the specific characteristics of karst aquifers(i.e. fast infiltration, movement of water in cave systems) and tocalculate the contribution of the karst areas to streamflow asspring flow, the SWAT model structure (Fig. 3a) was adapted asshown in Fig. 3b.

This new SWAT Model structure (Fig. 3b) was the result of thecombination of two main studies: Baffaut and Benson (2009) andNikolaidis et al. (2013) and represents the concept of the karstmodel and its hydrological pathways. Hereafter, this model config-uration is called KSWAT and combines an adapted SWAT model anda karst-flow model.

In particular, the adapted SWAT model considers fast infiltrationthrough caves and sinkholes up to the deep-aquifer, while thekarst-flow model represents the fast movement of water in subter-ranean conduits and the lower movement in narrow fracturesthrough the interconnection of reservoirs structure.

The adapted SWAT model consists in representing sinkholes bywetlands with small drainage area and a large hydraulic conductiv-ity at the bottom of the wetlands and loosing streams were repre-sented by tributary channels with high hydraulic conductivity inthe stream bed.

All the percolation in the soil profile, stream losses and seepagefrom the bottom of wetlands directly recharges the deep aquifer(DA_RCHRG, mm H2O). This is achieved by setting the deep aquiferpercolation fraction (RCHRG_DP) to 1, imposing minimum ground-water delay (GW_DELAY equal to 1) and setting the groundwatercoefficient of capillarity rise (‘‘revap”) 0.1 in order to avoid thatwater moves from shallow into the overlying unsaturated zonewhile the baseflow from shallow and deep aquifers was imposednegligible (Table 2).

The spring’s discharge was simulated with a karst-flow modeldeveloped in ‘‘Excel” environment by Nikolaidis et al. (2013). The

Table 2Parameters and parameter ranges used in the calibration (in alphabetic order), in the adapted SWAT model and in the karst-flow model. In column ‘‘Process” the information aboutthe order and group of calibration processes: 1, snow process; 2, runoff process; 3, lateral flow process; 4: groundwater process; 5: karst process.

Model Parameter Description Process Range ofcalibrationvalues

Range of values inkarst-subbasins

Karst-flow model a1 Fraction of inflow to upper reservoir 5 0–1Karst-flow model a2 Fraction of upper reservoir to lower 5 0–1SWAT/adapted

SWAT modelALPHA_BF Baseflow recession constant [1/d] 4/5 0–1 0.1

SWAT ALPHA_BF_D Baseflow recession constant for deep aquifer [1/d] 4 0 0Adapted SWAT

modelCH_K (1) Effective hydraulic conductivity in tributary channel alluvium

[mm h�1]5 – 300

Adapted SWATmodel

CH_K (2) Effective hydraulic conductivity in the main channel [mm h�1] 5 – 5–500 a

SWAT CN2 SCS runoff curve number for moisture condition II 2 �15% to +15% –SWAT EPCO Plant evaporation compensation factor 3 0.01–1 –SWAT ESCO Soil evaporation compensation factor 3 0.01–1 –SWAT/Adapted

SWAT modelGW_DELAY Groundwater delay [d] 4/5 0–500 1

SWAT/AdaptedSWAT model

GW_REVAP Groundwater ‘revap’ coefficient 4/5 0.02–2 0.1

SWAT GWQMN Threshold depth of water in the shallow aquifer required for returnflow to occur [mm]

4 0–1000 –

Karst-flow model kl Upper reservoir recession constant [1/d] 5 0–5 –Karst-flow model ku Lower reservoir recession constant [1/d] 5 0–5 –SWAT PLAPS Precipitation laps rate [mm/km] 1 0–100 –Karst-flow model Qk0 Initial karstic flow [m3/day] 5 0–3 � 105 –SWAT/adapted

SWAT modelRCHRG_DP Groundwater recharge to deep aquifer [fr] 4/5 0–1 1

SWAT REVAPMN Threshold depth of water in the shallow aquifer for revap to occur[mm]

4 0–500 –

SWAT SFTMP Snowfall temperature [�C] 1 �5 to +5 –SWAT SMFMN Minimum melt rate for snow on Dec 21 [mm �C�1 d�1] 1 0–10 –SWAT SMFMX Minimum melt rate for snow on Jun 21 [mm �C�1 d�1] 1 0–10 –SWAT SMTMP Snow melt base temperature 1 �5 to +5 –SWAT SNOEB Initial snow water content [mm] 1 0 –SWAT SOL_AWC Available water capacity of the soil layer [fr] 3 �25% to +25% –SWAT SOL_K Saturated hydraulic conductivity [mm h�1] 3 �25% to +25% –SWAT TIMP Snow pack temperature lag factor 1 0.01–1 –SWAT TLAPS Temperature laps rate [�C/km] 1 �10 to 0 –Adapted SWAT

modelWET_FR Fraction of subbasin area that drain into wetlands 5 – 0.1–1b

Adapted SWATmodel

WET_K Hydraulic conductivity through bottom of wetland [mm h�1] 5 – 300b

Adapted SWATmodel

WET_NSA,WET_MXSA

Surface area of wetlands at normal water level and at maximumwater level [ha]

5 – 20b

Adapted SWATmodel

WET_NVOL,WET_MXVOL,WET_VOL

Volume of water stored in wetlands when filled to normal,maximum water level and initial volume [104 m3]

5 – 300b

a CH-K (2) was manually calibrated in the final calibration.b Wetlands parameters were setting based on geological map extension.

A. Malagò et al. / Journal of Hydrology 540 (2016) 64–81 69

input was the daily deep aquifer recharge (DA_RCHRG, mm) calcu-lated by the SWAT model.

The karst-flow model is composed by an upper reservoir withfaster response that represents wide conduits in a karst systemand a lower reservoir with a slower response that simulates nar-row fractures (Kourgialas et al., 2010).

The karst-flow model uses the deep aquifer recharge of the sur-rounding subbasins and involved at the same time the simulationand calibration of springs discharges. The procedure required thecontribution/exclusion of deep aquifer recharge of nearest sub-basins and the manually adjustment of five parameters of karst-flow model minimizing the Normalized Root Mean Square Errors.

These five parameters included: Qko, the initial karst flow(m3/day), a1 the fraction of deep groundwater discharge enteringthe upper reservoir, a2 the fraction of flow from the upper reservoirdischarge entering the lower reservoir and ku and kl were recessionconstants (1/d) for the upper and lower reservoirs (Table 2). Forfurther details of karst-flow model equations, the reader can referto Nikolaidis et al. (2013) who provided a complete description.

The outputs of the karst-flow model were then introduced inSWAT as point sources with aggregated values at monthly timestep.

This modified SWAT was applied only in the karst-subbasinsidentified by three dominant soils that were strongly correlatedto the karst areas of the Geological Map (Fig. 4b and c). Accordingto the European Soil Database these are Leptosols (European SoilPortal, 2014), poorly developed and shallow soils over hard rockand comprise of very gravelly or highly calcareous material.

However, the use of the ‘‘dominant karst soils” sometimes waslimiting since in some cases the size of the subbasins (about20 km2) and the fragmentation of soils inside the subbasin didnot allow identifying it as ‘‘karst subbasin”. This occurred forinstance in the south-central part of the Island in Kourtaliotis Basin(corresponding to Kourtaliotis river basin in Fig. 1c), where thislimitation was overcome using literature information (i.e.Steiakakis et al., 2011) and the geological map. Thus, both karst-soils extension and the geological map were used to identifythe karst-subbasins where the KSWAT model was applied. It is

Fig. 4. Rivers and subbasins (a); definition of karst geological features: karst areas from geological map (b) and the subbasins with karst soils (karst subbasins) (c).

70 A. Malagò et al. / Journal of Hydrology 540 (2016) 64–81

noteworthy that the extension of wetlands inside each karst-subbasin was defined using the geological map, generally coveringthe entire subbasins where the karst-soils are dominant, or only apercentage of subbasin area as in the case of the Kourtaliotis Basin.The resulted area covered by the karst-subbasins was estimatedaround 2600 km2.

2.4.3. Model setupThe SWAT model requires pedological, climatological, topo-

graphical and land use data. Subbasins were delineated using theArcSWAT interface with a Digital Elevation Model of 25 m pixelsize (EU-DEM; Bashfield and Keim, 2011). The Digital ElevationMap was obtained from a Pan-European elevation data at 1 arc-second (EU-DEM). Subbasins and streams were defined using adrainage area threshold of 1000 ha resulting in 352 subbasins withan average area of 19 km2 covering 6700 km2 (Fig. 4a).

Land cover was derived from a 1 km raster map built from thecombination of CAPRI (Britz, 2004), SAGE (Monfreda et al., 2008),HYDE 3 (Klein and van Drecht, 2006) and GLC (Bartholome andBelward, 2005) for the year 2005. Land use was obtained fromthe Agriculture statistics of Greece (2005). For each subbasin the‘‘dominant land cover approach” was used in order to reduce thecomplexity and the computational level of the model. Sixclasses of dominant land cover were defined: arable land, pasture,

forest, urban area, water and range grasses. Each subbasin withdominant arable land cover was further subdivided in three parts(15–35–50% of subbasin area) in order to distribute more accu-rately the crops reported by the Agricultural census. Soil typeand characteristics were defined using a 1 km soil raster map,obtained from the Harmonized World Soil Database (HWSD)(FAO, 2008).

From the combination of land use, dominant soils, and singleslope for each subbasin the final configuration resulted in 502HRUs (Hydrological Response Units) with an average area of13 km2. The management practices for each crop included plant-ing, fertilization, irrigation and harvesting. The simulated timingof plant sowing and harvesting were implemented in SWATthrough daily heat unit concept. In this study the heat units foreach crop were calculated by Bouraoui and Aloe (2007) using thePHU (Potential Heat Units) program (PHU, 2007), developed atTexas Agricultural Experiment Station. The PHU program calcu-lates the total number of heat units required to bring a crop tomaturity. For each crop the crop growing season (winter versusspring crop), the base growing temperature (�C), the optimumgrowing temperature (�C), the dry down fraction, the time to matu-rity (number of days between planting and harvesting) were con-sidered as input in the PHU program. These attributes wererelated to the different European climatic zone.

A. Malagò et al. / Journal of Hydrology 540 (2016) 64–81 71

The amount of manure and mineral fertilization applied wereretrieved from the Common Agricultural Policy RegionalizedImpact (CAPRI) agro-economic model (Britz and Witzke, 2008). Itprovided at NUTS 2 administrative level (Nomenclature of Territo-rial Units for Statistics) the mineral and organic nitrogen and phos-phorus fertilizer application estimated from sectoral statistics andanimal production levels.

A manual irrigation was applied selecting the irrigated HRUs fora total irrigated area around 25,000 ha in Chania, 9300 ha inRethymno, 60,000 ha in Heraklion and 26,000 ha in Lasithi. Thetotal abstractions for irrigation use in each Prefecture was89 Mm3/y in Chania, 35 Mm3/y in Rethymno, 170 Mm3/y in Herak-lion and 66 Mm3/y in Lasithi, respectively. Both irrigated areas andvolumes were obtained from the Agriculture statistics of Greece(2005).

According to the National statistics, the sources of abstractionfor irrigation uses were set from deep aquifer and shallow aquifer,and from springs. In the model the abstraction for irrigation fromsprings (38 Mm3/y) was implemented as outside-watershed sourcein the schedule management plan and the corresponding abstrac-tions were subtracted from the daily deep aquifer recharge(DA_RCHRG, m3/s) (outside SWAT), so that the calibration ofsprings discharge in the karst-flow model was performed usingthe net of the abstractions.

The abstractions for drinking water, industry and olive mills(40 Mm3/y) were introduced in SWAT by subtracting water fromthe deep aquifer and shallow aquifer at monthly time step, whilethe abstractions for drinking water from the springs (24 Mm3/y)were subtracted from the daily deep aquifer recharge (DA_RCHRG)(outside SWAT).

The climate data used in this study include 69 stations withdaily data for precipitation and 21 stations for temperature from1961 to 2009. Monthly statistics of solar radiation, wind speedand relative humidity were calculated using the pan Europeanhigh-resolution gridded daily data set EFAS-METEO, acronym ofEuropean Flood Awareness System – METEO (Ntegeka et al.,2012) for 29 stations uniformly distributed over the modeledterritory.

The subbasins were split into elevation bands, and snow coverand snowmelt were simulated separately for each elevation band.In this study steep subbasins were subdivided into 4 classes ofelevation.

2.4.4. Calibration of streamflowFor the calibration and validation of the Crete SWAT model 22

stations of streamflow were used. Based on the simplest test tech-nique (Klemeš, 1986), 15 of them were used for calibration and 7for validation covering the period 1980–2009. The selected sub-basins for calibration were headwaters and were uniformly dis-tributed in the Island. The streamflow was calibrated in the 15selected subbasins in different steps (‘‘step-wise calibration”)according to the involved hydrological process, providing a cali-brated parameter set NOP (Near Optimal Parameters set). Theobtained NOP set was then transposed in ungauged subbasinsusing the ‘‘hydrological similarity approach”.

The step-wise calibration is a procedure that allowed calibrat-ing the different components of the streamflow according to differ-ent hydrologic processes underpinning each calibration objective.This procedure proposed by Pagliero et al. (2015) and Malagòet al. (2015) was performed using the SWAT-CUP (SWAT Calibra-tion Uncertainty Procedures) programs and the SUFI-2 algorithm(Abbaspour, 2008).

The calibration period covered 30 years from 1980 to 2009 with10 years of warm up. Among several parameters in SWAT, 19parameters (see parameters that refer to SWAT and SWAT/AdaptedSWAT model in Table 2) were chosen for the streamflow calibration

as the most sensitive and representative of each hydrologicalprocess involved in the calibration. The parameters and their initialranges (Table 2) were selected based on preliminary model runs,literature reviews (van Griensven at al., 2006) and sensitivity test.It is noteworthy however that the parameters in karst-subbasinswere not changed keeping the settings of the adapted SWAT modelconfiguration as described in Table 2.

In this work, since the streamflow data was only available atmonthly resolution, the streamflow was not divided into the maincomponents (surface runoff, lateral flow, baseflow). However, thecalibration was still performed step wise:

Step 1: Calibration of total monthly flow in order to control thetiming of runoff signal by adjusting the snow parametersby minimizing the coefficient of determination r2;

Step 2: Calibration of total monthly flow using only the parame-ters that regulate the surface runoff process by minimizingthe coefficient of determination r2;

Step 3: Calibration of total monthly flow using only the parame-ters that regulate the lateral flow process by minimizingthe coefficient of determination r2;

Step 4: Calibration of total monthly flow using only the parame-ters that regulate the baseflow process by minimizingthe Nash Sutcliffe Efficiency, NSE (Nash and Sutcliffe,1970);

Step 5: Final step, calibration of total monthly flow using all previ-ous parameters together with a reduced uncertaintyranges by minimizing the NSE coefficient.

The coefficient of determination r2 (Taylor, 1990) was adoptedas objective function for the first three steps and the NSE coeffi-cient (Nash and Sutcliffe, 1970) was selected for the fourth andfinal step as described below. The selection of the objective func-tions is the result of several tests since the differences betweenobjective functions on parameter sets could be quite striking(Abbaspour et al., 2015). The calibration was performed using1000 simulation runs at each step.

2.4.5. The regionalization and classification of ungauged subbasinsThe Final Near Optimal Parameter set (NOP) of calibrated sub-

basins (called ‘‘donors”) was transposed to ungauged subbasinsusing the hydrological similarity approach. The measure of hydro-logic similarity between subbasins is translated into subbasin char-acteristics which are common for gauged and ungaugedwatersheds. These characteristics that linked the subbasin charac-teristics with the hydrological responses have been performed inthe literature (Shamir et al., 2005; Oudin et al., 2010). However,for a comprehensive enlightenment of hydrological similarityapproach the reader can refer to Wagener et al. (2007) andSawicz et al. (2011). The calibration for the parameters of thekarst-subbasins were kept unchanged using the settings of theadapted SWAT model.

The hydrological similarity approach was performed using thePLSR (Partial Least Squares Regression) that allowed identifyingsimilar subbasins based on the correlation between the watershedcharacteristics and the discharge characteristics. Following Malagòet al. (2015), this correlation was performed using 17 gauged inde-pendent basins using the R package ‘‘pls” (Mevic and Wehrens,2007). The gauging stations selected to perform the PLSR analysisinvolved stations from the group selected for the calibration andsome from the validation group in order to identify a group of sta-tions with almost 8 years of continuous data from 1988 to 1995that allowed calculating robust statistical indices.

For the selected subbasins 20 independent variables represent-ing the subbasin characteristics (‘‘c” matrix) and 2 dependent flowvariables (‘‘q” matrix) were used in the PLSR analysis as the

72 A. Malagò et al. / Journal of Hydrology 540 (2016) 64–81

responses and regressors (q � c) (Malagò et al., 2015). The inde-pendent variables included river length (km), minimum, maximumand average elevation (m), median slope (%), content of clay andsand in the soil, percentage of different classes of land cover (%),percentage of karst area, mean annual precipitation (mm), mini-mum, maximum and average temperature (�C), annual potentialevapotranspiration (mm) and average number of days with precip-itation. The dependent variables included the mean annual dis-charge (mm/y) and the corresponding coefficient of variation.The PLSR analysis defined latent variables that were used toperform the classification of subbasins.

Table 3Comparison of observed and simulated streamflow in the period 1983–2009 after the intrname of the stations; in the third column the drain area (km2); the % data represents the % oand simulated volumes are respectively Mm3/year OBS and SIM; the performance indicatorsquare error (NRMSE%), the percent bias (PBIAS%), the Nash Sutcliffe Coefficient (NSE), the cslope of the regression line between simulated and observed values (br2).

Station Name Drain area (km2) % data Mm3/year OBS

STR32 Koiliaris 132.6 21 180.97STR19R Platis 209.6 69 50.15STR7R AnapodiarisH 522.7 54 30.13STR13 Giofiros 186.3 69 22.24STR20 Prasanos 101.9 54 17.83STR9R Geropotamos 394.5 69 17.21STR3 Katavothres 21.35 54 16.63STR26 Kakodikianos 78.11 77 15.89STR27 Sebreniotis 28.48 80 15.83STR11 Anapodiaris Pla 89.51 47 11.89STR10 Koutsoulidis 132.2 54 11.53STR2R Mirtos 96.34 84 10.87STR6R Aposelemis 204.8 69 10.69STR28 Rumatianos 22.06 76 6.75STR8 Litheos 41.98 77 6.39STR14R Gazanos 186.8 54 6.32STR1 Patelis 84.57 56 5.91STR4 Kalamafkianos 36.14 10 5.41STR15 Baritis 105.5 44 4.08STR25 Agios Vasiliou 35.64 74 2.37STR16 Arvis 26.53 80 1.66STR12R Iniotis 105.9 44 1.56

R Gauging stations used in the validation of regionalization technique.

Fig. 5. Some examples, comparison between monthly simulated and observed stream

Due to the limited number of potential donors (15 subbasins, 4%of total number of subbasins), it was necessary to define ‘‘hydro-logical regions” and then the ‘‘classification” of similar subbasinswas performed inside each region. For a review of regionalizationtechniques, the reader can refer to Parajka et al. (2005).

Using the latent variables, the regionalization was performedwith the Ward’s hierarchical cluster analysis (Ward, 1963) basedon the Euclidean distances and the classification of subbasins indonor’s classes was pursued inside each hydrological regionaccording to the K-Nearest-Neighbor using the R package ‘‘sup-clust” (Dettling and Maechler, 2012). The classification of each sub-

oduction of the springs (complete validation): in the first two columns the code andf observed monthly values in the period 1983–2009 (324 total months); the observeds between monthly observed and simulated steamflow are the normalized root meanoefficient of determination (r2) and the coefficient of determination multiplied by the

Mm3/year SIM NRMSE% PBIAS% NSE r2 br2

182.74 16.5 3.3 0.68 0.69 0.6246.30 11.1 2.4 0.63 0.63 0.4731.55 6.9 16.5 0.78 0.8 0.7322.94 12.5 25.4 0.53 0.66 0.6522.17 13.2 2.8 0.23 0.36 0.2733.04 32.2 183.4 �3.94 0.46 0.2411.56 11 �32.7 0.53 0.62 0.319.02 11.6 31.6 �0.87 0.25 0.2312.45 8.5 �17.9 0.58 0.6 0.3812.91 13.4 �2 0.64 0.65 0.5313.07 5.7 17.2 0.83 0.84 0.8211.31 15.3 39.9 �0.35 0.29 0.2626.46 14.4 158.7 �0.05 0.63 0.57.18 16.3 36.9 �0.06 0.26 0.186.13 20.6 8.2 �0.31 0.26 0.2

11.86 36.9 111.8 �3.05 0.23 0.176.12 12.1 13.7 0.48 0.57 0.513.57 40.7 10.4 �1.33 0.32 0.33.06 13.2 61.7 �0.04 0.19 0.124.62 28 103.4 �2.38 0.26 0.21.14 11.1 11.2 0 0.32 0.232.21 23.7 57.3 �0.81 0.16 0.12

flow values from 1983 to 2009 after the allocation of springs in the SWAT model.

Fig. 6. Cumulative frequency curves of percent bias (PBIAS, %) obtained betweenmonthly and observed streamflow before (only adapting SWATmodel) and after theintroduction of springs (KSWAT).

A. Malagò et al. / Journal of Hydrology 540 (2016) 64–81 73

basin allowed the automatic assignation of the NOP of ‘‘donor sub-basins” to the classified ungauged subbasins.

In order to verify the strength of the parameters regionalizationin the ungauged subbasins the model performance was evaluatedusing the percent bias (PBIAS; Gupta et al., 1999) between monthlyobserved and simulated streamflow at seven gauging stations. Thiswas an intermediate validation that anticipated a final completevalidation performed in all gauging stations after the introductionof springs in the model as described in the next paragraphs.

2.4.6. The calibration of springs and delineation of karst recharge areaof each spring

The springs were calibrated using the daily recharge of deepaquifer (DA_RCHRG) from the adapted SWAT model, except forSpring SP54 (Almiros Heraklion) that was calibrated usingDA_RCHRG at monthly time step. The springs were calibrated man-ually changing parameters of the karst-flow model in reasonableranges: the initial karst flow (m3/day), the fractions a1, a2, andthe recession constants ku and kl (1/d) for each reservoir as showin Table 2. This process ended when the Normalized Root Mean

Fig. 7. Some examples, comparison between daily simulated and observe

Square Error (NRMSE) calculated using the simulated and observedvalues for each spring was minimized.

42 springs out of 47 were calibrated having hypothesized thatsome springs drained the same karst area. This assumption wasstrongly correlated to the neighborhood location of these springsthat didn’t allow distinguishing the contribution of deep aquiferrecharge from the same karst-subbasins; hence SP12 and SP13,SP44 and SP45, SP4 and SP50, SP7, SP18 and SP19 were groupedinto SP12-13, SP44-45, SP18-19-7 and SP4-50.

The springs were entered as points sources in the model and theassociated calibrated monthly Qk (m3/s) were implemented asinput data in SWAT. For each spring the karst recharge areas(inside and outside the hydrological boundary) was defined assum of subbasins areas whose deep aquifer recharge contributesto a specific spring.

The subbasins that contributed to spring flow through theirdeep aquifer recharge in the karst-flow model were not necessarilyhydrologically connected. The linkage between these subbasinsand the springs represented the movement of water through thecave systems whose spatial extent and organization was differentfrom the surface subbasin hydrologic boundaries.

The performances of the karst-flow model were evaluated byvisual comparison of measured and simulated hydrographs ofsprings and using performance indicators including the Normal-ized Root Mean Square Error (NRMSE), the percent bias (PBIAS;Gupta et al., 1999), the coefficient of determination r2 multipliedby the slope of the regression line between simulated and observedvalues (br2; Krause et al., 2005) to overcome the intrinsic weaknessof each indicator (Bennett et al., 2013).

2.4.7. Final calibration and complete validationAfter the introduction of calibrated springs in the SWAT model

as point sources, an additional manual calibration was necessary toadjust the water discharge including peaks and baseflow since theaddition of the spring contribution modified the already calibrateddischarge. In particular, the effective hydraulic conductivity inmain channel (mm/hr) in Messera, Anapodiaris and in PetrasBasins was adjusted (Table 2).

d spring’s values from 1983 to 2009 (monthly comparison for SP54).

74 A. Malagò et al. / Journal of Hydrology 540 (2016) 64–81

Finally, the complete SWAT model was validated using thewhole dataset of streamflow gauging stations in order to evaluatethe model performance after the introduction of the springs. Sim-ilarly to calibrated streamflow and spring’s discharges, the com-plete SWAT model was validated comparing monthly observedand simulated streamflow in the period 1983–2009 calculatingthe NRMSE (%), PBIAS (%), the Nash Sutcliffe efficiency (NSE), br2

and r2.

3. Results and discussion

3.1. Hydrological simulation

The performance indicators obtained comparing the monthlyobserved and simulated streamflow for the 22 stream gauging sta-tions after calibration, regionalization and the introduction of thecalibrated springs are summarized in Table 3.

It is noteworthy that about 64% of the calibrated gaugingstreamflow stations reached satisfactory PBIAS% (values in the

Table 4Comparison of observed and simulated spring discharges in the period 1983–2009: in thecontributes of each spring (km2) and involves the area inside the hydrological basin and ouperiod 1983–2009 (324 total months); the observed and simulated volumes are respecobserved and simulated spring discharges are the normalized root mean square error (NRMdetermination (r2) and the coefficient of determination multiplied by the slope of the reg

Station Name Estimated karst recharge area(km2)

SP54a Almiros-Heraklion 323.48SP41 Stylos 185.01SP9 Almiros-Agios-Nikolaos 155.42SP43 Platanos/kalamionas-Agia 110.18SP32 Kourtaliotis 103.37SP42 Meskla 114.2SP40 Kourbos 53.32SP38 Vrisses 20.21SP39 Armenoi 78.48SP37 Petres 26.2SP44_45 Elliniki-Konto Kinigi 40.32SP34 Spilianos 34.07SP36 Mousela Xalikouti 47.95SP46 Drapania 40.9SP48 Therisos 12.6SP28 Geropotamos 25.58SP2 Zakros 28.45SP3 Chochlakies 34.75SP27 Panormos Almiro Nero 16.74SP18_19_7 Kria Vrisi-Simis-Kefalovrisi

Viannou35.14

SP22 Fodele 22.45SP52 Argiroupoli 24.23SP49 Kefalovris/Kalamoukas-Ierapetra 17.91SP12_13 Almiros Mallia-Grammatikaki 22.89SP6 Archon Stavrochoriou – Sitia 28.91SP20 Zaros – Votomos 26.38SP30 Spili 14.21SP8 Kalo Chorio 14.61SP11 Zou 29.54SP47 Sfinari 32.57SP31 Agia Fotia-Spili 14.21SP25 Emparos 16.14SP26 Seises 2.5SP4_50 Agios Georgios Sitia-Sikia Sitia 18.27SP21 Gergeri 41.8SP35 Ligres 8.35SP5 Lithines-Sitia 18.73SP14 Avli Migilisi 18.85SP51 Loutraki 33.22SP10 Petikou 3.22SP16 Mega Vrisi 28.97SP53 Agios Georgios-Viannou 3.69

a For SP54 the performance indicators were calculated using monthly values.

range ± 25%), while only 40% had NSE greater than 0.5 (seeMoriasi et al., 2007 for performance rating as reference). Further-more, 50% and 36% of calibrated streamflow gauging stationsreached respectively r2 and br2 larger than 0.5.

Fig. 5 shows some examples of comparison of hydrographs.Even though some of the statistical indicators did not score veryhigh, the hydrographs comparison illustrates that the model wasable to reproduce accurately the monthly variations over a longperiod of simulation even in subbasins influenced by excessivewater pumping, such as STR16.

This gauging station was fed by precipitation and discharge ofspring SP16. After the year 1990, the spring’s discharge wasdiverted to supply domestic uses reducing drastically the stream-flow in particular during the summer and autumn.

It was estimated that the percentage of streamflow gauging sta-tions that reached satisfactory PBIAS (PBIAS < ±25%, Moriasi et al.,2007) increased from 59% to 64% adding spring’s discharges inthe SWAT model highlighting the significant influence of springsin the water balance of Crete. Fig. 6 shows the PBIAS cumulativecurves of performance indicators calculated between monthly

first two columns the code and name of the stations; in the third the karst area thattside; the % data represents the percentage of observed values (once a month) in the

tively Mm3/year OBS and SIM; the performance indicators calculated between dailySE%), the percent bias (PBIAS%), the Nash Sutcliffe Coefficient (NSE), the coefficient ofression line between simulated and observed values (br2).

%data

Mm3/yearOBS

Mm3/yearSIM

NRMSE%

PBIAS%

NSE r2 br2

77 235.72 236.59 9.8 0.7 0.77 0.8 0.879 85.81 168.85 66 102.7 �4.8 0.64 0.3180 83.22 83.73 15.9 3.9 0.22 0.54 0.5610 67.11 67.43 23.8 6.1 0.18 0.55 0.5984 37.99 37.08 9.9 �0.8 0.44 0.43 0.4584 30.7 30.94 17.4 11.2 0.31 0.32 0.3868 29.85 28.68 14.6 2.1 �0.61 0.12 0.1277 26.95 26.53 13.4 �3 0.48 0.41 0.5184 24.79 24.09 26.1 1.2 �0.08 0.5 0.570 13.77 13.76 15.2 2 0.34 0.26 0.3984 13.5 13.52 9.5 5.3 0.44 0.32 0.4584 12.65 11.24 10.7 �6.3 0.36 0.21 0.3677 11.95 11.26 9.4 18.9 0.24 0.36 0.4175 9.87 6.35 14.9 �21.9 0.16 0.21 0.3242 9.68 9.35 16.6 �3 0.32 0.18 0.3264 7.75 7.81 14.2 �2.5 0.13 0.13 0.2280 5.4 5.77 9.5 �1.8 0.62 0.59 0.6349 5.33 5.47 17.3 2.3 0.34 0.38 0.4669 4.68 4.65 13.1 4 0.07 0.49 0.584 4.65 4.74 20.6 26.7 0.17 0.55 0.67

84 3.78 3.99 8.7 �3.2 0.21 0.13 0.2244 3.67 3.54 18.7 18.7 0.09 0.51 0.684 3.54 3.59 10 �4.8 0.41 0.33 0.4680 3.51 3.43 12.7 10.4 0.49 0.48 0.5683 3.26 3.23 12.5 �1.7 0.38 0.49 0.5110 3.15 3.03 32.4 �0.1 �0.06 0.04 0.0421 3.07 3.1 30.4 �2 �0.85 0.1 0.1176 3.05 3.1 14.3 1.2 0.29 0.26 0.348 2.14 2.14 13.2 �14.7 0.51 0.35 0.5284 1.96 1.99 12.8 �0.2 0.42 0.43 0.4935 1.94 1.97 21.4 �9.7 0.29 0.26 0.3636 1.53 1.52 13 11.5 �0.11 0.01 0.0280 1.53 1.23 16.7 �20.9 0.05 0.3 0.442 1.52 1.49 23.8 103.4 �1.06 0.26 0.3517 1.49 1.86 14.8 �8.3 0.57 0.57 0.6317 1.46 1.41 14.7 14 0.41 0.46 0.4781 1.38 1.31 11.9 �3.6 0.54 0.42 0.564 1.15 1.02 50.7 43.8 �1.04 0.41 0.6258 1.09 1.19 19.1 27.2 �0.1 0.42 0.4980 1.01 1.02 8.1 11.5 0.62 0.56 0.6334 0.65 0.64 20.3 32.7 0.04 0.1 0.1633 0.11 0.2 19.1 7 0.22 0.37 0.39

A. Malagò et al. / Journal of Hydrology 540 (2016) 64–81 75

and observed streamflow before and after the introduction ofsprings as point sources in the SWAT model.

In particular, streamflow discharge of gauging stations STR32(Koiliaris), STR16 (Avli), and STR10 (Koutsoulidis) were stronglyinfluenced by the discharge of springs SP41 (Stylos), SP14-SP25(Migilisi and Emparos) and SP20 (Zaros), respectively. For STR32the PBIAS decreased to 3% and the NSE increased to 0.68, forSTR16 the PBIAS% reached 11%, and finally for STR10 the PBIAS%and NSE were also satisfactory reaching 17% and 0.83 respectively(Table 3).

These results demonstrated that the karst-flow model correctlysimulated the discharge of springs increasing the SWATmodel per-formance. Fig. 7 shows some examples of comparison betweendaily observed and simulated spring flows (only for SP54 monthlycomparison), and Table 4 summarizes the comparison betweenmonthly and observed long-term annual discharge on the period1983–2009 and the performance indicators for all springs. It isnoteworthy that for more than 70% of the calibrated springs thePBIAS% is within the range of ±25%, indicating that the perfor-mance of the karst-flow model was satisfactory (Moriasi et al.,2007). NSE was positive for 80% of calibrated springs, while r2

and br2 had a wide range of values with around 50% of the valueslarger than 0.4.

However, the karst-flow model markedly overestimated the dis-charge for springs SP4-50 (Agios Georgios Sitia-Sikia Sitia), SP14(Avli Migilisi), SP16 (Mega Vrisi), SP41 (Stylos) and SP51 (Loutraki)maybe due to the quality of the observed data.

Concerning SP41, the most studied spring in Crete (Nikolaidiset al., 2013; Moraetis et al., 2010; Kourgialas et al., 2010), wasintentionally calibrated to overestimate the flow in order to takeinto account the contribution of ungauged springs. Nikolaidiset al. (2013) reported that Stylos spring has two permanent springsat elevation +17 m AMSL (with 85 Mm3/year of volume), but therewas also an ungauged intermittent spring (Anavreti at elevation+24 m AMSL). Both contributed to Koiliaris discharge. The authorsargued that these springs have a total average discharge around154 Mm3/y (2007–2010), and this volume was in agreement withthe total overestimated volume of Stylos in this work (150Mm3/y,for the same period 2007–2010, 170 Mm3/y considering 1983–2009). Finally, it is noteworthy that the spring SP46 (Fig. 7) hadtwo different hydrological behaviors: observed data were higherprior to year 1988, then the values decreased sharply highlighting

Fig. 8. Examples of estimated karst r

changes in the regime of discharge. This may be explained by achange in the method of measurement or more probably by anexcessive pumping of water as occurred for the STR16 spring.

3.2. The estimated recharge areas of karst springs

Fig. 8 shows some schematic examples of recharge areas ofsprings obtained considering the subbasins that contribute withtheir deep aquifer recharge to the calibration of springs. Table 4summarized also the karst recharge areas (km2) for each springor group of springs. The karst recharge areas of springs coverkarst-subbasins and karst areas for a total of 1928 km2, about the70% of total karst areas in Crete. The main karst recharge areadrains more than 300 km2 into Almiros Heraklion Spring (SP54),followed by the karst recharge area of Almiros-Agios-NikolaosSpring (SP9) that cover about 155 km2, and 130 km2 of karstrecharge area of springs in Koiliaris Basin.

To the best of our knowledge this work was the first estimationof karst recharge areas of all gauged springs in the Island of Crete,providing valuable information for water resources management.The only exception was the karst area drained by Almiros Springand the knowledge of extended areas (outside the hydrologicalboundary of basin) of springs in Koiliaris Basin.

Bonacci and Fistanic (2006) pointed out the difficulties to definethe karst area drained by Almiros Spring due to complex geologyand hydrogeology. There were different assumptions of its bound-aries and size which varies from 300 km2 (Arfib et al., 2000) to500 km2 (Lambrakis et al., 2000). However, Bonacci (1995) andBonacci and Ljubenkov (2005) estimated the karst recharge areaof Almiros of about 300 km2, similar to our finding.

The area outside the Koiliaris Basin that contributed to the karstrecharge area of Stylos Spring and other ungauged springs in Koil-iaris Basin was estimated around 80 km2 (7 subbasins that con-tribute with 70% of their total area). As a consequence, morethan 60% of karst recharge area of Koiliaris springs was outsidethe boundary of the hydrological basin. This result was confirmedby Nikolaidis et al. (2013) that estimated an extended karst area ofsprings in Koiliaris Basin of 79 km2 based on hydrologic modelingand geologic cross-sections.

Albeit this approach may be criticized by the fact that karstrecharge areas should be only delineated based on hydrogeologicalconsiderations derived, i.e. by tracer tests, it has to be taken into

echarge area of selected springs.

Fig. 9. Maps of monthly variations of precipitation, evapotranspiration and the deep aquifer recharge (Mm3/month).

76 A. Malagò et al. / Journal of Hydrology 540 (2016) 64–81

account that there are limits in the use of tracer studies in karstareas such as in Koiliaris basin. In a geological karst area which dis-charges a large volume of water there is a risk of high degree of

dilution of tracers (Knithakis, 1995) which in turn limits the use-fulness of the methods in delineating the extend area. In addition,breakthrough curves are highly dependent on the flow conditions,

Fig. 10. Annual water balance in different hydrological conditions (in percentage), average of 27 years (total volume of precipitation 6370 Mm3/y), normal year (total volumeof precipitation 6400 Mm3/y), dry year (total volume of precipitation 3700 Mm3/y) and wet year (total volume of precipitation 9600 Mm3/y). ET, evapotranspiration; SURQN,surface runoff excluding losses from tributaries and wetlands; LATQ, lateral flow; GWQ1, baseflow; DA_RCHRGN, amount of water recharge in the deep aquifer without thecontribution for springs that discharge inside (Qkin) and outside (Qkout) Crete Watershed; SA, shallow aquifer storage.

A. Malagò et al. / Journal of Hydrology 540 (2016) 64–81 77

often necessitating the repetition of the tracing experiment underlow flow and high flow conditions as observed by Goldscheider(2005) and Ravbar et al. (2011).

3.3. Spatial and temporal variation of hydrological components inCrete

The mean annual precipitation ranges from 550 mm (driestyear) to 1400 mm (wettest year) during the period 1983–2009,with a decreasing gradient from west to east. The precipitationwas higher at high elevation, reaching more than 2500 mm inthe wet years (1996, 1997, 2001 and 2003) and around 1000 mmin the central north areas, while in the plain areas and along thesouth cost the average annual rainfall was below 1000 mm.

The mean calculated actual evapotranspiration of Crete variedfrom 340 mm/y (driest year) to 390 mm/y (wettest year) reachingthe maximum values long the north coast and, in particular in Kar-teros, Aposelemis, Tzermiadon, Myrtos and Anapodiaris basins(east-central part of Crete).

Large values of surface runoff were simulated in the westernpart of Crete and in particular in Tavronitis Basin with maximumvalue in wet years, such as in 1997 and 2003 (more than1100 mm). Minor contributors to the water balance include lateralflow and baseflow. The baseflow reached high values in Kourtalio-tis, Akoumianos and Platis basins (south-central Crete). The annualand spatial variations of deep aquifer recharge were strongly corre-lated to the precipitation, as a consequence in zones with high pre-cipitation the deep aquifer recharge reached the largest valuessince the main mountainous systems were associated with themost important karst systems, allowing water to penetrate directlyto the deep aquifer as showed in Fig. 9. In particular, Fig. 9 illus-trates how precipitation, evapotranspiration and recharge of thedeep aquifer change during the year. Precipitation and deep aquiferrecharge changed during the year decreasing from January (�1120and 463 Mm3/month respectively) to August (�18 and 62 Mm3/month respectively) and increasing from September (�163 and63 Mm3/month respectively) to December (�1250 and 416 Mm3/month respectively). During the wettest months, the west part ofCrete reached the highest values of precipitation and also the high-est value of deep aquifer recharge. The evapotranspirationincreased from January (�240 Mm3/month) to April (�320 Mm3/month), in particular in subbasins long the cost, and thendecreased until August (�54 Mm3/month) and increased fromSeptember (�71 Mm3/month) to December (�240 Mm3/month).These long-term spatially and monthly variations could be very

useful for planning and implementing conservations measuresand programs and evaluating their performance.

3.4. The estimated water balance of Crete

Chartzoulakis et al. (2001) and many other studies (RCG, 2002;Vardavas et al., 2004; MEDIWAT, 2013; Koutroulis et al., 2013;Baltas and Tzoraki, 2013) pointed out the importance of the evap-otranspiration on the water balance of the whole Crete giving infil-tration a secondary role. Conversely to these findings, the CreteSWAT model has allowed the estimation of the water balance ofCrete resulting in significantly different estimates. Fig. 10 summa-rizes the percentage of flow contribution of the water balance maincomponents with respect to the long-term average volumes anddifferent hydrological conditions during the period 1983–2009.In addition, Fig. 11 represents the previous results following theKSWAT model approach providing a more readable and systematicinformation of the fluxes.

The evapotranspiration volume ranged from around 2250 Mm3/yin the driest year to 2500 Mm3/y in the wettest year exhibiting asmall variation in terms of absolute values but with a significantvariation with respect to the total volume of precipitation(Fig. 10). In the wettest year the main component of hydrologicalprocesses was the deep aquifer recharge, while in the driest yearthe evapotranspiration had the main role. From dry to wet yearthe surface runoff increased from 95 Mm3/y to more than800 Mm3/y and the total net water available in the deep aquiferincreased from 550 Mm3/y to around 3500 Mm3/y in dry and wetyear, respectively.

As a consequence, during the wet conditions there was highinfiltration, but also the surface runoff was larger than that duringdriest and normal hydrological condition. On the other hand, in dryyears there was also significant percolation to the deep aquifer(1000 Mm3/y). This result was also highlighted by Hartmannet al. (2014, 2015) pointing out that karst regions might be moreresilient to climate change in terms of both flooding and droughts.Furthermore, the authors demonstrated the existence a nonlinearrelationship between precipitation and recharge rate indicatingthat the recharge of deep aquifer is more sensitive to a decreasethan to an increase of precipitation (Hartmann et al., 2014). Thiswas confirmed also in this study where the total deep aquiferrecharge was estimated 28% of precipitation in driest year, whilewas 50% in wettest year departing of 16% and 6% from normal con-ditions, respectively.

Concerning the percentage of water balance components(Fig. 11), it was estimated that from the total precipitation of

Fig. 11. Annual water balance schematization in different hydrological conditions, dry (a), normal (b) and wet (c) year. ET, evapotranspiration; INF, infiltration in the soil;SURQN, surface runoff excluding the losses from tributaries and wetlands; LATQ, lateral flow; GWQ1, baseflow; DA_RCHRGN, amount of water recharge in the deep aquiferwithout the contribution for springs that discharge inside (Qkin) and outside (Qkout) Crete; WYLDk, freshwater contribution; SA, shallow aquifer storage.

78 A. Malagò et al. / Journal of Hydrology 540 (2016) 64–81

6400 Mm3/y about 40% was lost through evapotranspiration, 55%infiltrates and only 5% was surface runoff. The surface runoffincluding lateral flow (LATQ), baseflow (GWQ1) and spring’s dis-charge (Qkin) contribution to streamflow represents 22% of totalprecipitation. The total amount of deep aquifer recharge was

around 44% of which 14% contributed as springs discharge tostreamflow (Qin = 9%) or sea (Qout = 5%) (Fig. 11b). These percent-ages were valid for all years in normal hydrological conditions(for instance year 1987), but they changed in extreme hydrologicalconditions.

A. Malagò et al. / Journal of Hydrology 540 (2016) 64–81 79

Considering the driest year (year 1990, Fig. 11a), it was esti-mated that from the total precipitations of 3700 Mm3/y about60% was lost through evapotranspiration, 37% infiltrated and only3% was surface runoff.

The amount of deep aquifer recharge decreased with respect tothe normal hydrological condition to 28% of which 13% contributedas springs discharge to streamflow (Qin = 9%) or sea (Qout = 4%). Thetotal contribution to freshwater decreased to 14% with respect tothe normal condition.

In the wettest year (year 2003, Fig. 11c), considering a totalvolume of precipitation around 9600 Mm3/y, evapotranspirationwas estimated around 26%, 65% was infiltration and 9% was sur-face runoff. The deep aquifer recharge reached the highest per-centage of 50% and the water yield increased to 28%. Our workhas shown that the evapotranspiration ranged between2250 Mm3/y in a driest year and of 2500 Mm3/y in wettest year,corresponding to 60% and 26% of total precipitation, respectively.These results suggested the importance of karst water supplyduring dry periods.

3.5. Strengths and challenges of the karst modeling approach

The modeling approach proposed in this study allowed achiev-ing satisfactory simulations of streamflow and springs dischargesat monthly and daily time step respectively across the whole CreteIsland. These temporal resolutions are inherently linked to theavailable input data (Table 1), the nature of the simulated hydro-logic processes and the modeling objectives (Baffaut et al., 2015).Conversely to previously studies that focus mainly on annual orlong-term mean annual prediction (i.e. Malard et al., 2015), herethe monthly simulations of streamflow have allowed capturingthe seasonal variations and gathering precious information for abetter water resources management. Similarly, the daily simulatedsprings discharges provided information of each karst rechargearea and the spring hydrograph, if further investigated, can alsoexpress some geometrical and hydraulic characteristics of aquifers(Fiorillo, 2014).

Furthermore, the spatial and temporal (annual and monthly)distribution of simulated water resources (i.e. precipitation, deepaquifer recharge, evapotranspiration and runoff) and the meanannual water balances in different hydrological conditions areimportant benefits of the present work.

Despite these strengths, the modeling approach suffers fromsome limits. For instance, the monthly time step of streamflowhas limited the application of step-wise calibration not allowinga reliable estimation of each component of the streamflow. Fur-thermore, in data scarce areas, where only annual time step isavailable and spatial data has coarse resolutions, the reproducibil-ity of the KSWAT model needs some adaptations that should beevaluated case by case. Likewise, the knowledge of springs dis-charges and their position is vital and in extremis they could beapproximated using literature information. In this context, softdata has an important role and should be seen as injections ofsome insightful common sense into the automatic or semiauto-matic calibration procedure (Seibert and McDonnell, 2002). Forinstance, Malard et al. (2015) used 42 alpine and peri-alpinespecific springs discharges from literature as references sites inSwitzerland for karst recharge assessments, and Hartmann et al.(2015) used 40 independent observations of mean annualrecharge both from field and modeling studies over Europe andMediterranean.

Finally, albeit the identification of karst recharge areas of eachspring is a valuable aspect of the KSWAT model, the associateduncertainty was not object of this study and this topic deservesfurther investigations that may involve climatic and topographicdescriptors as pointed out in Hartmann et al. (2015).

4. Conclusion

In this study the SWAT model was adapted and integrated witha karst-flow model to simulate the karst water resources in theIsland of Crete. The KSWAT model allowed calibrating the dailyspring discharges in 47 gauging stations, estimating their karstrecharge areas and their introduction in the SWATmodel increasedthe performance of streamflow prediction in karst-subbasins.These springs contributed significantly to total discharge with300 Mm3/y in dry hydrological condition, ensuring ecological ser-vices (Malard et al., 2015), and 850 Mm3/y in wet periods. In addi-tion, the seasonal variation of volume of springs suggests thatthese valuable sources should be conserved and preserved in par-ticular from April to September when available volumes are thelowest and agriculture and tourism demand increases.

The analysis of the water balance also showed that waterresources are not homogeneously distributed in Crete and changesignificantly in different hydrological conditions. In particular,the western part of Crete has a surplus of water resources withrespect to the eastern part where there is low water availabilityand high demand. Messara Valley (Geropotamos Basin) is an exam-ple of intensively managed basin in south-eastern Crete where anoverexploitation of groundwater is occurring in a large number ofwells since 1984 that continue to decrease groundwater levels atalarming proportions (Kritsotakis and Tsanis, 2009).

In this context, the present study provides a methodology and atool for the integrated water management of Crete and other sim-ilar areas, by providing detailed spatially distributed hydrologicbalances and accurate estimation of water availability using bothhard and soft data.

Acknowledgements

The authors are grateful to the reviewers for their preciousremarks, which made it possible to improve the quality of thepaper. Funding for the work was provided by the EU FP7-ENV-2009 Project SoilTrEC ‘‘Soil Transformations in European Catch-ments” (Grant# 244118). The work of Dionissios Efstathiou hasbeen co-financed by the European Union (European Social Fund –ESF) and Greek national funds through the Operational ProgramEducation and Lifelong Learning of the National Strategic ReferenceFramework (NSRF) – Research Funding Program: Heracleitus II.Investing in knowledge society through the European Social Fund.The shapefiles of monitoring points of streamflow and springs, aswell as the databases of the related time series are published onResearchGate.

References

Abbaspour, K.C., 2008. SWAT-CUP2: SWAT Calibration and Uncertainty Programs –A User Manual. Department of Systems Analysis, Integrated Assessment andModelling (SIAM), Eawag, Swiss Federal Institute of Aquatic Science andTechnology, Duebendorf, Switzerland.

Abbaspour, K.C., Rouholahnejad, E., Vaghefa, S., Srinivasan, R., Yang, H., Kløve, B.,2015. A continental-scale hydrology and water quality model for Europe:Calibration and uncertainty of a high-resolution large-scale SWAT model. J.Hydrol. 524 (1), 733–752.

Agriculture Statistics of Greece, 2005. PEIPAIATM-2009-PIREAS.Arfib, B., de Marsily, G., Ganoulis, J., 2000. Pollution by seawater intrusion into a

karst system: new research in the case of the Almyros source (Heraklion, Crete,Greece). Acta Carsologica 29 (1), 15–31.

Arnold, J.G., Youssef, M.A., Yen, H., White, M.J., Sheshukov, A.Y., Sadeghi, A.M.,Moriasi, D.N., Steiner, J.L., Amatya, D.M., Skaggs, R.W., Haney, E.B., Jeong, J.,Arabi, M., Gowda, P.H., 2015. Hydrological processes and model representation:impact of soft data on calibration. Trans. ASABE 58 (6), 1637–1660. http://dx.doi.org/10.13031/trans.58.10726.

Arnold, J.G., Srinivasan, R., Muttiah, R.S., Williams, J.R., 1998. Large area hydrologicmodeling and assessment: Part I. Model development. J. Am. Water Resour.Assoc. 34 (1), 73–89.

80 A. Malagò et al. / Journal of Hydrology 540 (2016) 64–81

Baffaut, C., Benson, V.W., 2009. Modeling flow and pollutant transport in a karstwatershed with SWAT. Trans. ASABE 52, 469–479.

Baffaut, C., Dabney, S.M., Smolen, M.D., Youssef, M.A., Bonta, J.V., Chu, M.L., Guzman,J.A., Shedekar, V., Jha, M.K., Arnold, J.G., 2015. Hydrologic and water qualitymodeling: spatial and temporal considerations. Trans. ASABE 58 (6), 1661–1680. http://dx.doi.org/10.13031/trans.58.10714.

Bakalowicz, 2015. Karst and karst groundwater resources in the Mediterranean.Environ. Earth Sci. 74, 5–14.

Baltas, E., Tzoraki, O., 2013. Water Resources Management on the Island of Crete:Lessons Learnt. In: United Nations International Year of Water Cooperation(Eds.), Free Flow – Reaching Water Security Through Cooperation, Tudor Rose.p. 285–9.

Bartholome, E., Belward, A.S., 2005. GLC2000: A new approach to global land covermapping from Earth observation data. Int. J. Remote Sens. 26 (9), 1959–1977.

Bashfield, A., Keim, A., 2011. Continent-wide DEM creation for the European Union.In: 34th International Symposium on Remote Sensing of the Environment,Sydney, Australia, 10–15 April 2011 <http://www.isprs.org/proceedings/2011/ISRSE-34/211104015Final00143.pdf> (accessed 21st January 2015); dataavailable at <http://www.eea.europa.eu/data-and-maps/data/eu-dem#tab-metadata> (accessed 21st January 2015).

Bennett, N.D., Croke, B.F.W., Guariso, G., Guillaume, J.H.A., Hamilon, S.H., Jakeman,A.J., Marsili_libeli, S., Newham, L.T.H., Norton, J.P., Perrin, C., Pierce, S.A., Robson,B., Voinov, A.A., Fath, B.D., 2013. Characterising performance of environmentalmodels. Environ. Model. Softw. 40, 1–20.

Bonacci, O., 1995. Brackish karst spring Pantan. Acta Carsologica 24, 97–107.Bonacci, O., Fistanic I., 2006. Contribution to hydrological analysis of the coastal

karst spring Almyros (Creete, Greece). In: Conference onWater Observation andinformation system for decision support, Balwois (Ohrid, Republic ofMacedonia). Radic (Eds.), UNDP-Agency of Skopje. p. 191–2.

Bonacci, O., Ljubenkov, I., 2005. Nova saznanja o hidrologiji rijeke Krke. HrvatskeVode 13 (52), 265–281.

Bouraoui, F., Aloe, A., 2007. European Agrochemicals Geospatial Loss Estimator:Model development and Applications, EUR – Scientific and Technical ResearchSeries, ISSN 1018-5593. Office for Official Publications of the EuropeanCommunities, Luxembourg.

Britz, W., 2004. CAPRI Modeling System Documentation. Final Report of the FP5Shared Cost Project CAP-STRAT ‘‘Common Agricultural Policy Strategy forRegions, Agriculture and Trade.” QLTR-2000-00394. Univ. Bonn, Bonn,Germany.

Britz, W., Witzke, P., 2008. CAPRI Model Documentation 2008: Version 2. In: Britz,W., Witzke, P. (Eds.). Bonn: University of Bonn, Institute for Food and ResourceEconomics.

Buchanan, T.J., Somers, W.P., 1976. Discharge measurements at gaging stations.Techniques of Water-Resources Investigations of the United States GeologicalSurvey. Washington, D.C.: USGS. Book 3, Chapter A8.

Chartzoulakis, K.S., Paranychianakis, N.V., Angelakis, A.N., 2001. Water resourcesmanagement in the island of Crete, Greece, with emphasis on the agriculturaluse. Water Pol. 3 (3), 193–205.

Daly, D., Dassargues, A., Drew, D., Dunne, S., Goldscheider, N., Neale, S., Popescu, C.,Zwhalen, F., 2002. Main concepts of the ‘‘European Approach” for (karst)groundwater vulnerability assessment and mapping. Hydrogeol. J. 10 (2), 340–345.

Dettling, M., Maechler, M., 2012. Supclust: Supervised clustering of predictorvariables such as genes. R package version 1.0-7.

EC, 2003. Vulnerability and risk mapping for the protection of carbonate (Karst)aquifers 2003. In: Zwahlen, F. (Ed.), COST-620 Action Final Report. EuropeanCommission, DG-Research, p. 320.

EC, 2004. The main coastal karstic aquifers of southern Europe 2004. In: Calaforra, J.M., (Ed.), COST-621 Action, Groundwater Management of Coastal KarsticAquifers, European Commission, DG-Research, 126.

EU-DEM, Metadata <http://www.eea.europa.eu/data-and-maps/data/eu-dem#tab-metadata>.

European Soil Portal, 2014 <http://eusoils.jrc.ec.europa.eu/ESDB Archive/ESDB/>(accessed April 2014).

FAO, 2008. Harmonized World Soil Database (version 1.0). FAO, Rome. Italy andIIASA, Laxenburg, Austria.

Fiorillo, F., 2014. The recession of spring hydrographs, focused on karst aquifers.Water Resour. Manage. 28, 1781–1805.

Goldscheider, N., 2005. Fold structure and underground drainage pattern in thealpine karst system Hochifen-Gottesacker. Eclogae Geol. Helv. 98 (1), 1–17.http://dx.doi.org/10.1007/s00015-005-1143-z.

Gupta, H.V., Sorooshian, S., Yapo, P.O., 1999. Status of automatic calibration forhydrologic models: comparison with multilevel expert calibration. J. Hydrol.Eng. 4 (2), 135–143.

Hartmann, A., Gleeson, T., Rosolem, R., Pianosi, F., Wada, Y., Wagener, T., 2015. Alarge-scale simulation model to assess karstic groundwater recharge overEurope and the Mediterranean. Geosci. Model Dev. 8, 1729–1746. http://dx.doi.org/10.5194/gmd-8-1729-2015.

Hartmann, A., Mudarra, M., Andreo, B., Marín, A., Wagener, T., Lange, J., 2014.Modeling spatiotemporal impacts of hydroclimatic extremes on groundwaterrecharge at a Mediterranean karst aquifer. Water Resour. Res. 50 (8), 6507–6521. http://dx.doi.org/10.1002/2014WR015685.

Klein, G.K., van Drecht, G., 2006. HYDE 3: Current and historical population and landcover. In: Integrated Modeling of Global Environmental Change. An Overview ofIMAGE 2.4.93-111. Netherlands Environmental Assessment Agency (MNP),Bilthoven, The Netherlands.

Klemeš, V., 1986. Dilettantism in hydrology: transition or destiny. Water Resour.Res. 22 (9), 177S–188S.

Knithakis, M., 1995. Hydrogeological Investigation of Rethymnon Region(Prefecture of Rethymnon). Institute of Geological and MineralogicalInvestigations (IGME), Rethymnon.

Kourgialas, N.N., Karatzas, G.P., Nikolaidis, N.P., 2010. An integrated framework forthe hydrologic simulation of a complex geomorphological river basin. J. Hydrol.381, 308–321.

Koutroulis, A.G., Tsanis, I.K., Jacob, D., 2013. Impact of climate change on waterresources status: a case study for Crete Island, Greece. J. Hydrol. 479, 146–158.http://dx.doi.org/10.1016/j.jhydrol.2012.11.055.

Krause, P., Boyle, D.P., Base, F., 2005. Comparison of different efficiency criteria forhydrological model assessment. Adv. Geosci. 5, 89–97.

Kritsotakis, M., Tsanis, I.K., 2009. An integrated approach for sustainablewater resources management of Messara Basin, Crete, Greece. EWRA 27(28), 15–30.

Lambrakis, N., Andreou, A.S., Polydoropoulos, P., Georgopoulos, E., 2000. Nonlinearanalysis and forecasting of a brackish karstic spring. Water Resour. Res. 36 (4),875–884.

Malagò, A., Pagliero, L., Bouraoui, F., Franchini, M., 2015. Comparing calibratedparameter sets for the Scandinavian Peninsula and for the Iberian Peninsula.Hydrol. Sci. J. 60 (5), 949–967. http://dx.doi.org/10.1080/02626667.2014.97833.

Malard, A., Sinreich, M., Jeannin, P., 2015. A novel approach for estimating karstgroundwater recharge in mountainous regions and its application. Hydrol.Process. http://dx.doi.org/10.1002/hyp.10765.

MEDIWAT, 2013. State of the art of Water Resources in Mediterranean Islands.Available at: <http://www.mediwat.eu/sites/default/files/D.3.1.1.pdf>.

Mevic, B., Wehrens, R., 2007. The pls package: principal component analysis andpartial least squares regression in R. J. Stat. Softw. 18 (2), 1–23.

Monfreda, C., Ramankutty, N., Foley, J., 2008. Farming the planet: 2. Geographicdistribution of crop areas, yields, physiological types, and net primaryproduction in the year 2000. Global Biogeochem. Cycles 22, GB1022.

Moraetis, D., Efstathiou, D., Stamati, F., Tzoraki, O., Nikolaidis, N.P., Schnoor, J.L.,Vozinakis, K., 2010. High frequency monitoring for the identification ofhydrological and biochemical processes in a Mediterranean river basin. J.Hydrol. 389, 127–136.

Moriasi, D.N., Arnold, J.G., Van Liew, M.W., Bingner, R.L., Harmel, R.D., Veith, T.L.,2007. Model evaluation guidelines for systematic quantification of accuracy inwatershed simulations. Trans. ASABE 50, 885–900.

Nash, J.E., Sutcliffe, J.V., 1970. River flow forecasting through conceptual models.Part I: a discussion of principles. J. Hydrol. 10, 282–290.

Neitsch, S.L., Arnold, J.G., Kiniry, J.R., Srinivasan, R., Williams, J.R., 2010. Soil andWater Assessment Tool Input/Output File Documentation Version 2009,Grassland. Soil and Water Research Laboratory, Agricultural Research Serviceand Blackland Research Center, Texas Agricultural Experiment Station, CollegeStation, Texas.

Nikolaidis, N.P., Bouraoui, F., Bidoglio, G., 2013. Hydrologic and geochemicalmodeling of a karstic Mediterranean watershed. J. Hydrol. 477, 129–138.

Ntegeka, V., Decloedt, L.C., Willems, P., Monbaliu, J., 2012. Quantifying the impact ofclimate change from inland, coastal and surface conditions. In: Klijn, F.,Schweckendiek, T. (Eds.), Comprehensive Flood Risk Management – Researchfor Policy and Practise, CRC Press, Taylor & Francis Group, Leiden, TheNetherlands; In: Proceedings of the 2nd European Conference on Flood RiskManagement FLOODRisk 2012, Rotterdam, The Netherlands, 19–23 November2012.

Oudin, L., Kay, A., Andréassian, V., Perrin, C., 2010. Are seemingly physically similarcatchments truly hydrologically similar? Water Resour. Res. 46, W11558.http://dx.doi.org/10.1029/2009WR008887.

Pagliero, L., Bouraoui, F., Willems, P., Diels, J., 2015. Large-scale hydrologicalsimulations using the soil water assessment tool, protocol development, andapplication in the Danube Basin. J. Environ. Qual. 43, 145–154.

Parajka, J., Merz, R., Bloschl, G., 2005. A comparison of regionalisation methods forcatchment model parameters. Hydrol. Earth Syst. Sci. 9, 157–171.

PHU 2007. POTENTIAL HEAT UNIT (PHU) PROGRAM download <http://swat.tamu.edu/software/potential-heat-unit-program/> (accessed 21st January 2007).

Rantz, S.E., et al., 1982. Measurement and Computation of Streamflow: Volume 1.Measurement of Stage and Discharge. U.S. Geological Survey Water-SupplyPaper 2175.

Ravbar, N., Engelhardt, I., Goldscheider, N., 2011. Anomalous behaviour of specificelectrical conductivity at a karst spring induced by variable catchmentboundaries: the case of the Podstenjšek spring, Slovenia. Hydrol. Process. 25(13), 2130–2140. http://dx.doi.org/10.1002/hyp.7966.

RGC, 2002. Integrated Water Resources Management on Crete. Study Made Underthe Department of Water Resources Management (in Greek).

Sawicz, K., Wagener, T., Sivapalan, M., Troch, P.A., Carillo, G., 2011. Catchmentclassification: empirical analysis of hydrologic similarity based on catchmentfunction in the eastern USA. Hydrol. Earth Syst. Sci. 15, 2895–2911.

Seibert, J., McDonnell, J.J., 2002. On the dialog between experimentalist andmodelerin catchment hydrology: use of soft data for multicriteria model calibration.Water Resour. Res. 38 (11), 23.1–23.14.

Shamir, E., Iman, B., Morin, E., Gupta, H.V., Sorooshian, S., 2005. The role ofhydrograph indices in parameter estimation of rainfall–runoff models. Hydrol.Process. 19, 2187–2207.

Smart, C.C., Worthington, S.R.H., 2004. Springs. In: Gunn, J. (Ed.), Encyclopedia ofCaves and Karst Science. Fitzroy Dearbon, London, pp. 699–703.

A. Malagò et al. / Journal of Hydrology 540 (2016) 64–81 81

Steiakakis, E., Monopolis, D., Vavadakis, D., Lambrakis, N., 2011. Effective infiltrationassessment in Kourtaliotis karstic basin (S. Crete). Adv. Res. Aquatic Environ.,485–492

Taylor, R., 1990. Interpretation of the correlation coefficient: a basic review. J. Diagn.Med. Sonog. (January/February), 35–39

Tzoraki, O., Nikolaidis, N.P., 2007. A generalized framework for modeling thehydrologic and biogeochemical response of a Mediterranean temporary riverbasin. J. Hydrol. 346, 112–121.

van Griensven, A., Meixner, T., Grunwald, S., Bishop, T., Diluzio, M., Srinivasan, R.,2006. A global sensitivity analysis tool for the parameters of multi-variablecatchment model. J. Hydrol. 324 (1–4), 10–23. http://dx.doi.org/10.1016/j.jhydrol.2005.09.008.

Vardavas, I., Chartzoulakis, K., Papamastorakis, D., Xepapadeas, A., Spanoudaki, K.,Zacharioudakis, G., Kritsotakis, M., Bertaki, M., 2004. Report of Water Resourcesof Crete, 2004. Technical Report.

Wagener, T., Sivapalan, M., Troch, P.A., Woods, R.A., 2007. Catchment classificationand hydrologic similarity. Geogr. Comp. 1 (4), 901–931.

Wang, X., Liang, Z., Wang, J., 2014. Simulation of runoff in Karst-influenced LianjiangWatershed using the SWAT Model. Sci. J. Earth Sci. 4 (2), 85–92.

Wang, Y., Brubaker, K., 2014. Implementing a nonlinear groundwater module in thesoil and water assessment tool (SWAT). Hydrol. Process. 28 (9), 3388–3403.

Ward, J.H., 1963. Hierarchical grouping to optimize an objective function. J. Am.Statist. Assoc. 58, 236–244.

Williams, P.W., 2004. Karst evaluation. In: Gunn, J. (Ed.), Encyclopedia of Caves andKarst Science, vol. 475. Fitzroy Dearbon, London, p. 8.

Wooldridge, S.A., Kalma, J.D., 2001. Regional-scale hydrological modelling usingmultiple-parameter landscape zones and a quasi-distributed water balancemodel. Hydrol. Earth Syst. Sci. Discuss. 5 (1), 59–74.

Yachtao, G.A., 2009. Modification of the SWAT Model to Simulate HydrologicProcesses in a Karst Influenced Watershed MS Thesis. Virginia-Tech,Department of Biosystems Engineering, Blacksburg.